Skip to content

[DeepseekV32]: use _concat_mla_absorb_q_general to replace torch.cat #12215

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Fridge003 merged 5 commits into from
Nov 12, 2025
Merged

[DeepseekV32]: use _concat_mla_absorb_q_general to replace torch.cat #12215

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
d927518
dsv32: use _concat_mla_absorb_q_general to replace torch.cat
bingps Oct 27, 2025
3859ef8
Merge branch 'main' into dsv32-q-cat
bingps Oct 28, 2025
ebbaa9a
Merge branch 'main' into dsv32-q-cat
bingps Oct 28, 2025
d6e8e1d
Merge branch 'main' into dsv32-q-cat
Fridge003 Nov 10, 2025
bce7513
Merge branch 'main' into dsv32-q-cat
Fridge003 Nov 12, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 7 additions & 6 deletions python/sglang/srt/layers/attention/nsa_backend.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,6 +21,7 @@
NSA_FUSE_TOPK,
compute_nsa_seqlens,
)
from sglang.srt.layers.attention.trtllm_mla_backend import _concat_mla_absorb_q_general
from sglang.srt.layers.dp_attention import get_attention_tp_size
from sglang.srt.model_executor.forward_batch_info import ForwardBatch, ForwardMode
from sglang.srt.utils import is_hip
Expand Down Expand Up @@ -911,7 +912,7 @@ def forward_extend(

if NSA_PREFILL_IMPL == "tilelang":
if q_rope is not None:
Copy link
Copy Markdown
Contributor

@gemini-code-assist gemini-code-assist bot Oct 27, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

critical

This change replaces torch.cat with _concat_mla_absorb_q_general. It's crucial to ensure that _concat_mla_absorb_q_general is functionally equivalent to torch.cat in all scenarios where it's being used. If _concat_mla_absorb_q_general does not handle all cases correctly, it could lead to incorrect results or unexpected behavior. Can you provide more details on the scenarios where _concat_mla_absorb_q_general is guaranteed to work correctly and any limitations it might have?

Given the potential for widespread impact, this is a critical issue.

                q_all = _concat_mla_absorb_q_general(q_nope, q_rope) # Ensure this function is equivalent to torch.cat in all cases

q_all = torch.cat([q_nope, q_rope], dim=-1)
q_all = _concat_mla_absorb_q_general(q_nope, q_rope)
return self._forward_tilelang(
q_all=q_all,
kv_cache=kv_cache,
Expand All @@ -921,7 +922,7 @@ def forward_extend(
)
elif NSA_PREFILL_IMPL == "flashmla_sparse":
if q_rope is not None:
q_all = torch.cat([q_nope, q_rope], dim=-1)
q_all = _concat_mla_absorb_q_general(q_nope, q_rope)

# NSA_FLASHMLA_BACKEND_DECODE_COMPUTE_FP8 has no effect here,
# because the flashmla_sparse kernel doesn't support fp8 compute
Expand All @@ -947,7 +948,7 @@ def forward_extend(
)
elif NSA_PREFILL_IMPL == "flashmla_kv":
if q_rope is not None:
q_all = torch.cat([q_nope, q_rope], dim=-1)
q_all = _concat_mla_absorb_q_general(q_nope, q_rope)
return self._forward_flashmla_kv(
q_all=q_all,
kv_cache=kv_cache,
Expand Down Expand Up @@ -1031,7 +1032,7 @@ def forward_decode(

if NSA_DECODE_IMPL == "flashmla_sparse":
if q_rope is not None:
q_all = torch.cat([q_nope, q_rope], dim=-1)
q_all = _concat_mla_absorb_q_general(q_nope, q_rope)
return self._forward_flashmla_sparse(
q_all=q_all,
kv_cache=kv_cache,
Expand All @@ -1041,7 +1042,7 @@ def forward_decode(
)
elif NSA_DECODE_IMPL == "flashmla_kv":
if q_rope is not None:
q_all = torch.cat([q_nope, q_rope], dim=-1)
q_all = _concat_mla_absorb_q_general(q_nope, q_rope)
return self._forward_flashmla_kv(
q_all=q_all,
kv_cache=kv_cache,
Expand All @@ -1054,7 +1055,7 @@ def forward_decode(
)
elif NSA_DECODE_IMPL == "tilelang":
if q_rope is not None:
q_all = torch.cat([q_nope, q_rope], dim=-1)
q_all = _concat_mla_absorb_q_general(q_nope, q_rope)
return self._forward_tilelang(
q_all=q_all,
kv_cache=kv_cache,
Expand Down
Loading